Confluence 2.3’s major new feature was Massive, the Confluence clustering solution. However, turning Confluence into a clustered app introduced some fairly fundamental changes into our architecture. For example, since there may be many, separate Confluence Massive nodes on physically separated servers, we can no longer use the filesystem for storage and must instead push all of that information to the database.
And that, of course, impacts how plugins are handled. Prior to Confluence 2.3, uploaded plugins were stored in the $CONFLUENCE_HOME/plugins directory. Starting with Confluence 2.3, when plugins are uploaded, they are written directly to the database, in a table called PLUGINDATA. As each Confluence node starts up, the plugins from that table are copied out into the directory $CONFLUENCE_HOME/plugins-cache. The files are refreshed each time, so manually modifying this directory will have no effect.
When you need to install new plugins, you’ll need to use the web interface. Or check out the the Confluence Plugin Developer Kit, which lets you install and uninstall plugins from a maven commandline.
I hope this information helps avoid any confusion as people begin to port their plugins to Confluence 2.3. We’re sorry for any inconvenience this change may cause plugin developers.

Changes to plugin storage in Confluence 2.3