I started getting seriously into bike work circa 2002 and was lucky enough to come across Jobst Brandt's view of the topic around that time. Since then, most of which time I've spent working as a mechanic, I've observed that the Brandtian observations of the mechanical dynamics at hand are wholly correct, but I don't agree with him on what to do about it.
He posted a number of times on rec.bicycles.tech on this subject, material that can be found if you look, but there's an excerpt on sheldonbrown.com that I'll repost here:
My cranks get loose, quite quickly too; over about 10 miles or so from being solid to flopping about in the breeze. Any suggestions?
One or both of the cranks are ruined! Once ridden in the "floppy"
mode, the tapered square bore of the crank has been deformed and can
no longer be secured on a spindle. Install and properly tighten new
cranks on the spindle after greasing the tapered square ends of the
spindle. Proper tightness should be achieved with a torque wrench or
by a skilled hand.
The admonition not to grease the spindle finds life mainly in the
bicycle trade. When I discussed the "dry assembly" rule with crank
manufacturers, I discovered that they had warranty claims from
customers who split cranks. However, cranks cannot be split by
overtightening them. This can be proven by attempting to do so. An
M8x1 bolt is not strong enough to split a major-brand crank.
Failure from "over-tightening" is caused by repeated re-tightening of
properly installed cranks. In use, an aluminum crank squirms on its
taper, and because the retaining bolt prevents it from moving off the
taper, it elbows itself away from the bolt and up the taper ever so
slightly. The resulting loss of preload, after hard riding, can be
detected by how easily the bolt can be turned.
Loss of crank bolt preload is greater on left than right cranks,
because left cranks transmit torque and bending simultaneously while
right cranks transmit these forces separately. The left crank
transmits driving torque through the spindle to the right crank and
chainwheel while the right crank drives the chainwheel directly.
Besides that, the right crank transmits torque to the spindle only
when standing on both pedals. Doing this with the right foot forward
(goofy footed) is the only time the spindle transmits reverse torque.
Mechanics, unaware of why crank bolts lose preload (and commensurate
crank tightening), have re-tightened bolts until cranks split. No
warnings against re-tightening properly installed cranks are evident
although it is here where the warning should be directed rather than
Because friction plays no role in torque transmission, preload in the
press fit must be great enough to prevent elastic separation between
the crank and spindle under torque and bending. This means that no gap
should open between crank and spindle facets under forceful pedaling.
Crank bore failure occurs when the press fit is loose enough that a
gap opens between spindle and crank. Torque is transmitted by both the
leading and the trailing half of each facet, contact pressure
increasing and decreasing respectively. In the event of lift-off, the
entire force bears only on the leading edge of facets and causes
plastic deformation, causing the bore to take on a "pincushion" shape
(loose crank syndrome). Subsequent tightening of the retaining screw
cannot correct this because neither the retaining bolt nor crank is
strong enough to re-establish the square bore.
The claim that a greased spindle will enlarge the bore of a crank and
ultimately reduce chainwheel clearance is also specious, because the
crank cannot operate in a plastic stress level that would soon split
the crank in use. However, increased engagement depth (hole
enlargement) may occur without lubricant, because installation
friction could ream the hole.
With or without lubricant, in use, cranks will make metal-to-metal
contact with the spindle, causing fretting erosion of the steel
spindle for all but the lightest riders. Lubricating the spindle for
assembly assures a predictable press fit for a given torque. Without
lubrication, the press is unknown, and galling (aluminum transfer to
the steel spindle) may occur during assembly. After substantial use,
spindle facets may show rouge and erosion from aluminum oxide from the
crank, showing that lubricant was displaced.
Crank "dust caps" have the additional duty to retain loose crank
bolts. Because crank bolts lose preload in use, they can become loose
enough to subsequently unscrew and fall out if there is no cap. If
this occurs, loss of the screw will not be noticed until the crank
comes off, after the screw is gone.
In other posts he put forward the theory that Campagnolo was the originator of the dry taper rule in cycling, because they were having issues with their cranks splitting at the taper bore with repeat tightening. Light vintage racing cranks such as old Record are in fact quite prone to this with lubricated spindles and repeat tightening. It is far harder to get such a failure out of any but the very fanciest contemporary square taper cranks. Most are pretty chunky.
In my experience, it is exactly right that square taper cranks squirm on the taper. It is also exactly right that if you grease the taper from the beginning, torque the crank bolts properly, and leave them alone, it will be fine. I've been doing this on my own bikes for about 17 years now with zero issues ever. I use 44 Nm as a generic square taper crank bolt value.
However, where the Brandtian line falls apart is that in practice, mechanics need to be able to rapidly torque check everything on a bike. We do it not necessarily because, as Brandt suggests, we don't understand that cranks squirm, but because poor initial assembly is rampant and we have to be able to correct it in an expedient fashion. That means crank bolts throughout their lives are going to have mechanics repeatedly putting a wrench on and tightening them until they hit full torque. A dry taper getting this treatment tends to reach a bottoming point where it's not going to squirm up the spindle any further, and where re-torquing the bolt won't drive it up the spindle any further either. If the interfaces were more commonly lubricated, splitting the taper would be a more widespread issue.
In some sense the ideal thing would be for mechanics to remove cranks all the time, grease the spindles to help ensure, as Brandt pointed out, the press fit is coming out as expected for a given torque input, and somehow compel the owner to make sure the crank bolts are never tightened again. In reality it's just not practical, and so we're left with the somewhat flawed de facto system of dry tapers as the norm. Brandt is correct that some galling of the mating surfaces can occur without lubrication, but it's rarely a meaningful problem.
Also, the nature of a press fit is that there's no room in there, so there's little if any difference in function between heavy and light lubrication. A thin coating leftover from the factory or applied as a rust inhibitor is pretty close functionally to slathering it with grease.