Two SPI slaves on one bus occasionally drive MISO at the same time and corrupt transfers. Chip select timing looks right. What else causes this - tri-state delays, or a slave that never releases the bus?