...my import and export times have tripled if not worse with 3.5 RC...
Sorry for the delay in replying, I got busy with some stuff the last few days.
There is one issue in the 3.5 RC that would cause inflated import in place times for certain photos, though the worst I've seen is a 70% increase and it's usually much less and often none at all (and for moderately sized batches of images just copied from card/camera is all but impossible to detect due to disk caching). That specific issue is (will be) resolved in the 3.5 final version. If that's the issue, the other thing that would slow down drastically is "Read Metadata" and it would get better if you subsequently perform a save metadata operation*, I think.
I'm not sure what would cause the issue you're describing, though. It makes me wish I was more familiar with using xperf/kernrate and other windows profiling tools in the field.
It makes it sound to me like an interaction with some low level piece of software like a virus scanner, search indexer, or driver. That is, some piece of software that actually operates inside the Lightroom process either by being in the kernel or an UI extension that injects itself into the process.
If you bring up task manager and check the "Show Kernel Times" option in the view menu, is it the red or green line that spikes when the machine seizes? If it's red, we're looking for a kernel component like a file system filter (virus scanner, search indexer, rootkit, etc) or video driver (I seem to recall LR in prior versions getting into really bad interactions with NVIDIA's nView software) and if it's green then it's either LR itself or a bad interaction with some kind of extension. There are other more technical tools for drilling in further, but I won't elaborate on that here.
P.S. if it matters I am using dual displays. One is 1920x1080 and the other is 1680x1050 the graphics card is a NVIDIA GeForce GT 220.
Hmm. If you use LR on only one screen and/or disconnect the secondary one, does the manifestation change at all? That could be an interesting clue.
I wish I had a more slam dunk answer, but I am pretty curious to understand this problem. Chances are somebody somewhere has or will hit it, too.
Thanks -- DT
* The problem is due to an ACR feature that calculates an identifier for images. Once the id is saved into the XMP, it bypasses the expensive computation on subsequent uses of the photo. There are a bunch of cases that are inexpensive, but the fallback involves reading a few chunks of the actual image pixel data and running a digest algorithm on it.
