Here's where things get odd. I meant to try 3349 - WGS_1984_PDC_Mercator, but I made a typo and ended up putting 3394, which is apparently Nahrwan_1934_Iraq_Zone. But oddly enough, it's really close. I mean really. Around the US/Canada border the margin of error is about 5 Nautical Miles, up near 50N, it's only off by about 12 Nautical Miles. That's much better than the 60% margin of error I was seeing with 102100 which was listed as the WKID of Google Maps on multiple ArcGIS and ERSI pages. I'm a little concerned that an Iraq Zone projection from 1934 seems to be the closest spatial reference available, and I would hate for someone to attempt to buffer an area outside of what I have tested unaware that it may be nowhere near true. I'm hoping you can shed some light on the situation.
I realize that "measurements" of distance shouldn't be done on a Mercator, but as I am not measuring but simply buffering a line, I would assume that the ArcGIS server takes this into account to calculate the buffer when passed the proper (Web Mercator) spatial reference. So I remain confused as to why 102100 doesn't result in a proper buffer, and even more confused why a projection from Iraq in 1934 does.