let's be clear though that there will be plenty of capabilities that simply won't work. for example, you won't be able to accelerate data models. the 5.0 indexers simply won't know how to do that.
One BIG caveat here, though: a Splunk 6 search head will by default ask its distributed search peers to generate a remote timeline. This isn't a problem with 6.x indexers, but 5.x indexers won't know how to generate this and as a result searches might slow down dramatically.
The workaround is to set the following in limits.conf on the search head and restart Splunk:
[search] remote_timeline_fetchall = false
This can be removed when all indexers are upgraded to 6.x.