As a continuation of my previous post on this issue, further investigation of my XSLTTransformTimeout issues points to farm-level performance issues with the SQL cluster.
Since the last post and the corrective action of copying known “valid” XSL stylesheets to those frontend servers with mismatched files, the web part issue eventually recurred (just as we secretly expected). The next action plan from Microsoft was to increase the XSLTTransformTimeout value from 10 seconds to a whopping 30 seconds!
The farm-wide performance impacts aside, my admins caved and raised the timeout value. For about a week, things seemed to normalize and behave correctly… but then one morning, BAM! The web part issue returned, yet again.
So now with the timeout value at 30 seconds, XSL stylesheets consistent and manually verified across all the frontends, and the April 2014 CU applied, we were out of options.
The web part issue now ties directly back to larger performance issues on our farm, as a result of an overloaded SQL cluster. The only long-term resolution is to add additional SQL hardware (cue the bureaucratic red tape!).