Hi!
On a per-session basis, we cache the last heap block used for inserts and try to use the same block for subsequent inserts.
+1, from my POV this is good idea and it's cool that it already has implementation.
I'd suggest to do not 1\1 split for these pages, to keep tree lower. Like fillfactor/(1-fillfactor) split. Also, this will make splits less frequent.
Best regards, Andrey Borodin.