...which caused Gnash to not run under my qt4webkit-based browser.
I'm talking about NPPVpluginNeedsXEmbed value for NPP_GetValue(), which is supposed to be TRUE if plugin supports XEmbed. However, the corresponding piece of Gnash code is:

      case NPPVpluginNeedsXEmbed:
#ifdef HAVE_GTK2
          *static_cast<NPBool *>(aValue) = TRUE;
#else
          *static_cast<NPBool *>(aValue) = FALSE;
#endif
          break;

so for some reason it returns FALSE when GTK+ GUI is not compiled. AFAIK, this is blatant lies, since Gnash supported XEmbed from the beginning and doesn't even offer an alternative.
Anyways, some browsers like Firefox seem to sidestep this, but I use Qupzilla with Qt4 QtWebKit fork, and the latter aborts the init function if windowed plugin denies XEmbed support, leading to cascade of errors and the plugin not working. It's WebKit problem, so other browsers with the same engine are affected too, e.g. Arora.
So, in summary, Gnash didn't work under any Qt4 QtWebKit-based browser without the GTK+ GUI.
The most important point is that after changing the above FALSE to TRUE the plugin works just dandy. Thus, patch attached. I assume the whole #ifdef thing was some misunderstanding, it doesn't really make sense to me.