It's not recommended to access the same Sesame store from two different java applications directly. Sesame's SAIL (the database access layer) assumes it has sole control over the resources on disk, having two different JVMs apply two different SAIL objects to the same resources will likely lead to inconsistencies or deadlocks. So you should set things up via HTTP. Granted that this is less efficient, but Sesame has quite a few optimizations (custom binary serializations etc.) in place to speed up HTTP communication, so it should be quite workable.
An alternative is that you extend your Jersey app such that it exposes the (local) Sesame repository you are using with your own implementation of Sesame's REST protocol, including a SPARQL endpoint. If you do that, you can connect to your Sesame store from the Workbench without running it on a Sesame Server.
It's more work of course, but if performance of your own app is a big concern, this might be a good way to go. It's probably not as hard as it sounds since all the functional stuff (SPARQL engine, determining proper formats based on mime-types, etc. etc.) is all supplied by Sesame, you "just" need to tie it together.
But I'd give the first option (talking to a HTTP repository from your Jersey app) a try first, if I were you :)