GeoTools 2.6.x looks to be focused on quality and performance.
As always the scope and timing of releases is set by you - the members of our community. There is a host of interesting "unsupported" modules that can make it into this release based on time and funding.
We have tossed both gt-swing-widgets and gt-swing-widgets-pending in the bit bucket as their module maintainers have stepped down. We have salvaged a new gt-swing with s strict mandate to provide user interface classes for our demos; nothing too fancy - just enough that we are not wrecking out examples with user interface code.
We expect the render to gain additional capabilities as members of the public supply patches.
Using the dynamic symbolizers framework and the Eastwood charts project allows to paint charts over a map using the Google Charts syntax for defining the chart properties and data.
Process API improvements
The process API, still in unsupported land, has been changed to allow integration of external libraries such as Sextante as plugins of the GeoTools process API. No Sextante integration module is available due to licensing issues (Sextante is GPL) but you can find a bridge in the GeoServer experimental WPS module
Next Generation JDBC
Thus far the "Next Generation" JDBC DataStore are worth being excited about, they are more maintainable, with a cleaner architecture and more even support of advanced features (aggregation functions or support for dates for example). These are also available on the previous 2.5.x branch.
Andrea has organised a module clean up aligning modules with a module maintainer; or deleting modules that are unused as required.
Andrea Aime has been cleaning up one of the top issues on our technical debt page - cleaning up after the transition to GeoAPI filter done way back in GeoTools 2.3. The work is incomplete.
We have updated out data model to support the constructs used by the symbology encoding specification; and in a few isolated cases received patches to hook them up to the render (such as Unit of Measure support).
We also hope to see an introduction of a "N-Dimensional" GridCoverage "DataAccess" API. This work was started by Simone but does not appear to be finished in time for release.
There is a Technical Debt page with details of some "undone" work (often work left over when a commercial funding ended for a specific task).