Figures for 7.4.1:
     464,957,840 bytes allocated in the heap
     118,407,504 bytes copied during GC
      11,893,160 bytes maximum residency (9 sample(s))
       1,211,360 bytes maximum slop
              32 MB total memory in use (0 MB lost due to fragmentation)

For a version before my recent refactoring of TyClDecl/HsTyDefn
     472,031,360 bytes allocated in the heap
     122,203,224 bytes copied during GC
      12,964,944 bytes maximum residency (9 sample(s))
       1,209,688 bytes maximum slop
              34 MB total memory in use (0 MB lost due to fragmentation)

For HEAD:
     469,342,912 bytes allocated in the heap
     125,998,504 bytes copied during GC
      13,880,872 bytes maximum residency (9 sample(s))
       1,191,216 bytes maximum slop
              33 MB total memory in use (0 MB lost due to fragmentation)

So residency is up a bit, and I don't really know why but I think it's acceptable.