How can my Java/C++ hybrid applet retrieve data efficiently via HTTP with caching?

I'm converting a legacy C++ interactive Win32/OpenGL app into a browser plugin for Firefox and IE.  The initial plan was to use the NPAPI, then use the pluginhostctrl to wrap it as an ActiveX control for IE.  That plan fell flat because some features of the newer npruntime API were needed, and pluginhostctrl didn't support npruntime.  Then I looked at doing a native ActiveX version of the plugin, but the MS paradigm was too alien for me.  I ran away, screaming.

Now I've decided to try turning the plugin into a Java applet using JNI.  (Nice because it works in FF and IE.)  I've worked through the native window issues and figured out applet signing, but now I find myself stymied by the one piece of browser technology I thought was going to be most helpful: the asynchronous fetching and caching of files from a server.

The plugin needs to be able to retrieve and progressively display potentially large amounts of data retrievable in several-megabyte-sized units from an HTTP server.  It is important to be able to issue multiple simultaneous requests for data.  Likewise, it is important that client-side caching be employed, since data may be loaded into the display multiple times as it goes in and out of view.

My question is:  How can I implement the data retrieval conveniently, but without completely throwing performance out the window?  I can think of 3 options:  1) Use Javascript and XMLHttpRequest and the browser's built-in cache to get the data (but then how do I efficiently pass it to C++?); 2) Use Java and HTTPUrlConnection to get the data (but is it cached, and how do I efficiently pass it to C++?); or 3) Use C++ and something like curl to get the data (but then I have to implement my own cache).  Perhaps there is even a fourth option where C++ somehow talks directly to Javascript, if that's even possible.

Thanks!