Haven't used Java, but I have tried out Inline::Java (a year or two ago), which worked well. It also has a more recent release (although only 2011) and more good reviews, and Java seems to be failing all its tests on CPAN Testers (although Inline::Java doesn't look too great either)...

As for the approach in general - you'll lose some performance in the Perl<->Java interface. If your web interface doesn't get many queries that'll likely be fine, but for a heavily used interface you may run into performance issues, and a pure Java web interface would probably be advisable.

Not duplicating the DAOs is probably a good idea, especially if they contain some code (validation?), but if the schema is simple and you're willing to have duplicate DAOs, take a look at DBIx::Class.

Also, if you haven't yet, look at Dancer, Catalyst, Mojolicious, and Mojolicious::Lite.