You cannot design such stuff "on the limit". Mind the write latency of an sdcard as well.
4bit/SDIO mode - I did some testing with stm32f4discovery as it possess an SDIO interface. The r/w speed to an SDHC 4GB card (average one) was 4-5MBytes/sec when using 16-32kBytes blocks (mind that STM32F407 is an mcu with 168MHz clock, CortexM4, 192kB ram).