Came across this new blog post from Cormac Hogan (@vmwarestorage) who has written about a problem that EMC have discovered around SCSI UNMAP resize requests in vSphere 5 or 5.1.
Essentially there is a 2TB file limit in the latest VMware OS', and thus if you try and send a SCSI UNMAP command it will create a balloon file to enable it to move blocks and shrink the space. This balloon file cannot be bigger than the 2TB limit.
When using SCSI UNMAP you tell it to resize a percentage of the original volume size - so rather than telling the command to UNMAP a large percentage, it's recommended to perform a series of requests of smaller resize amounts to ensure you don't hit the 2TB ceiling.
Original blog post is available here: http://cormachogan.com/2013/05/14/heads-up-unmap-considerations-when-reclaiming-more-than-2tb/
Incidentally, i'd be interested to hear your views on SCSI UNMAP. Do you use it? If so, how often do you find you use it, and how much space do you reclaim (on average?)