The question was resolved by myself and no more help needed :)
I spent a night reconfiguring Fiasco and L4Linux and reading the Mailing list archive. And it appears that l4linux itself was configured correctly. But I encountered a troubles with 'Mysterious double fault' when using QEMU that could be overcome if disable 'Assembler IPC shortcut' option in Fiasco configuration. Also I found that options in loader config file have significant order of precedence. They should appear in the following order:
allow_cli allow_vga all_sects_writable
If all_sects_writable appears before allow_cli then loader says that l4linux task is not allowed to do any I/O. It is really strange and non-obvious. Also, some diversity in options order may lead to unresolved page faults and GPFs like the GPF which I was asking for.
Someone can say it is very trivial but I spent a night to figure out the way it works.
I just wanted to point these non-obvious moments to novices. (and, maybe, developers of loader server -- what if to make these options order to be non-significant? It could make loader behaviour more predictable.)
WBR, valery