I suggest to use the IB_Monitor to proof that the Schema-Caching is
responsible for weak performance. I use a simliar technique like you,
and i saw that the "create table" Statement sometimes need 1,6
seconds. The update Schema-Cache operation is "read only" and should
be quite quick.
See my log (http://orgamon.org/SQL.log.true.txt).

BUT - (jason on board?) if you have "CacheStatementHandles" set to
true, the Handle of the "cache-update" Statement is still valid, and
a "drop table " Statement will fail.

So my suggestion is to Jason (this will help all of us!): invent
a "Schema Cache Filter" that means: All Tables, with name
starts "TMP$" (this my be a user-property of the session) are ignored
by the schema cache dschingamading, this means they are invisible for
all the Schema-Cache Things - dont know if this will work, but ...

Another Suggestion: Invent a Schemacache.BeginUpdate;
SchemaCache.EndUpdate; Block, inside the Block a counter is
incremented id IBO wants to update the cache (but it don't) if at the
moment of EndUpdate the counter>0 than you may update.

Andreas