Consider the following hierarchy: there is an applet A plugged into Mozilla/Firefox browser. Inside the applet there is an AWT component C that implements XEmbed server specification (e. g. XEmbedCanvasPeer). Inside the component there is an XEmbed client S (e. g. StarOffice bean or XEmbeddedFrame).

Then, consider the current focused component is some HTML element, while A is not focused. If I click on some component inside A with mouse, it successfully gets focused, while if I click on S it doesn't. At the same time I can tab in and out of S from/to C.
The same problem may be reproduced in easier configuration: an applet with a single button. Set focus is on some HTML element other than applet, then click on some other window like terminal, and then click on a button inside the applet - the button doesn't get focused. Only the second mouse click makes it a focus owner.