Use four times more cache entries and divide the memory for each entry
by four. This lowers the linear read throughput somewhat but increases
the access speed for filesystems.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>