2 hour time window (inherited from BTC) seems waaaaay too lax.   Don't most (if not all) warp attacks stem from this laxity ?  
I think somebody looked into it, and propagation of data network wide takes an average of 6 seconds and maybe 12 seconds max, IIRC.  I understand the need to have some laxity in the ordering of timestamps, but ... 2 hours !?!  
Why not a time window of single-digit multiple of the block time?  That should be enough.... Maybe 3x or 5x the block time ?   With Verge's 30s block time, this window would then be 90s or maybe 150s (2m 30s). Blocks timestamped outside this window should be rejected.

Also, there ought to be a trust score on the peers a node maintains connections with, based on how long they've been connected and how closely a peer's idea of time matches the node's.  A peer should not be able to just appear out of the blue, and present a huge fork and have it accepted. Only if said peer has been around long enough to establish a min trust score should their blocks be accepted and re-broadcast.   Likewise, if a peers idea of time is more than 30 seconds or so off from ours, the trust score should be lowered (or denied entirely) until their clocks match reality.

What do people think ? Would this fix the exploit ?