I just checked with 1.4.1_02, and BufferedImages obtained through ImageIO still arn’t elligable for HW acceleration.
I still prefer them over toolkit.createImage though,
afterall, its not much work to copy them to an acceleratable BufferedImage.
Also, images obtained from toolkit.createImage have dodgy ImageProducers, that misbehave with some methods.
The most notable is image.createScaledInstance, which can generate partially corrupt images if called on images obtained from toolkit.createImage.
(I wonder if that bug has been reported…)
On the otherhand, the ImageIO.read() followed by copying it to an acceleratable Image may be slower.
I havn’t timed it.
Another issue that might be worth considering, is that ImageIO in 1.4.0 had a bug that prevented it from reading small image files :o
(I ran into that 1 with my 4k applet racing game :P)