I didn't have any problems with random crashing while converting the file outside Skyline, but I did find way to improve performance. I have tested with Thermo's newer file reading API and the SRM chromatogram access has a lot of potential for speedup. With the current one-transition-at-a-time enumeration, the new API is possibly faster than the old one, but not by much (and I haven't compared the actual chromatogram data points to make sure they're the same). But the new API also supports requesting multiple chromatograms at a time; when I request all 2800 transitions at once, it finishes in a mere 15 seconds. It also uses about 800MB of RAM.
Notwithstanding the extra RAM usage, it's probably worth it to switch to this all-transitions-at-once enumeration after ProteoWizard has switched to using the new Thermo API.