It's the first time I've read about it, but my initial understanding based on a cursory glance of the BIP is no.  Whichever Dandelion-compatible client is selected at random to actually broadcast the transaction, it strips out any of the extra information required by Dandelion and just sends a regular Bitcoin transaction.  Definitely slows down propagation a tiny bit, but doesn't appear to impact tx size once the final broadcast is made.  Clever stuff.

That said, it will obviously be slightly more resource intensive for those choosing to use Dandelion.  You'll be maintaining two distinct mempools.