Puts silently failing without error

I'm running membase on a windows machine (ubuntu linux in deployment, but haven't gotten there yet), developing in java, using spy memcached (not the pre-release version, intend to use moxi client-side for deployment). Membase has 256 megs of RAM on my dev machine. I need to ingest a couple of gigabytes of data, stored in ~4kb per key/value pair. After a couple of thousand stores, they start silently failing (= no error message, but data is not in membase upon later inspection). 

I've read the "Growing Data Sets Beyond Memory" info, and it would seem like I'm inserting stuff faster than membase can persist it to disk. That's ok, as long as my client code gets notified. On the GDSBM webpage it says that the "current recommendation" is to go with option 2 (return an error in this situation), but I'm not seeing any error code (the boolean future returned by spy memcached always returns true immediately upon get-ing it). 

Has anyone run into this issue and worked around it for java + spy memcached?