Recently, I found a text on the history of development of gear and bearing steels since 1940. In it, I found the answer to something from long ago.
In 1969, John Jacobson took me along to the Yamaha dealer show in Washington, D.C., where we hoped for a meeting with "Reo Rake" (Leo Lake), who was then Yamaha's racing manager. I did get a couple of minutes with him, during which he said that Yamaha big-end bearings had to be set up slightly loose, as the rollers swelled up slightly and permanently at first running. My eyes widened, but what did I know? Leo contrasted this with normal practice with roller big-ends in British Singles, which were always traditionally fitted to zero clearance.
What I read says that under the shear stress that results from compressive surface loading, any retained austenite in the roller material may spontaneously transition into martensite (these are crystal phases in the material). And the martensite is less dense than the austenite, so it has slightly greater volume.
During manufacture of the roller material, while it is extremely hot, most of it becomes austenite, and when this is quenched in heat treatment, a large part of the austenite transforms to the harder martensite. But not all! So there is the mechanism of Reo Rake's mysterious roller-swelling: conversion of retained austenite into martensite, driven by applied shear stress as the rollers carry the inertia loads of piston and rod, and combustion pressure.
—Kevin Cameron