How to use connect_max_errors to improve my cache service?

I used six membase servers to setup a cache cluster. Then I run a standalone moxi in another server to give cache proxy service. But I found when any of membase server down, my php may get empty value from the cache proxy. After 30 ~ 60 sec ( I set autofailover with 30 sec ) my php (with memcache protocal) can get correct value again. But 30 ~ 60 seconds seems too long for my services, it means my database will be directly read by client for a long time. Database may dead in the 30 seconds. But I found the following tips in wiki.

----------------------------------
[http://www.couchbase.org/wiki/display/membase/Follow+A+Request+Through+Moxi]

In the server-side moxi processes spawned by membase, the connect_max_errors is configured to be 5 and the connect_retry_interval is 30000 (that is, 30 seconds).
----------------------------------
I know there is a moxi process running on membase server.

Well, my question is:
in the 30 ~ 60 seconds, my php get empty value , inside of the membase, are there any error raised, or anything seem perfect?