I had an ah-ha moment last night. I forgot an important fact in this problem, SCK clock rate.
The easiest way to setup SPI on AVR is with a SCK rate of F_CPU/4. The first hacked Arduino software for SD used this rate. This library had to be customized for each use so it could fit the 168 that was then use on Arduino.
Adafruit's 4.7k/10k product sort of worked with this library.
Other people used the 1.8k/3.3k dividers that worked well at 4 Mhz
I decided to write my own FAT16 library that would be small enough for anyone to use and have some space for application software. I decided to use F_CPU/2. The Adafruit products didn't work so I did the 4050 buffer mod and convinced Adafruit to modify their product.
There were occasional problems with 1.8k/3.3k dividers. I modified the Fat16 library by adding an initialization argument to select F_CPU/2 or F_CPU/4.
I wrote SdFat when Arduino move to the 328. The Arduino company used this library in the SD.h wrapper but fixed the SCK speed at F_CPU/4. This avoided the problem with 1.8k/3.3k dividers.
Here is what my advice about dividers would be now:
Don't use dividers larger than 1.8k/3.3k.
If you have problems with 1.8k/3.3k dividers set SCK to 4 MHz or change them.
1k/2.2k dividers will work at 8 MHz even though they don't meet the 10 ns spec.
470/1k dividers should come close to meeting the 10 ns spec for 20 pf.
pito 200/410 dividers are as good as a buffer.
If you set the MOSI/SS/SCK pins to input after you finish with accessing the sdcard your power consumption might be even smaller with resistive divider smiley Worth of try..
You must keep chip select high or the card will not sleep. SCK is low when the card sleeps but MOSI is a problem in the current SdFat since I output 0XFF to put the card in sleep mode. I should try zero, then MOSI should be low.
I don't want to put SPI pins in input mode because it will cause libraries that share the bus with SdFat to fail. I already have problems with libraries that don't setup SPI before each bus access. I initialize the SPI bus before each transfer.
The 0.44 ma is total for everything on the SD module with 3.3V regulator. SD cards now sleep at less than 0.1 ma. some at less than 50 ua.
I checked other cards. 0.25 ma for the module with card sleeping is common for new cards. The module draws 0.18 ma when I pull the card.