It's about timing with pull-up resistors.
A large pull-up is going to recover slowly, because there's capacitance in the traces on the PCB (or the wires). If you "actually need" a pull-up, you don't want to go too high if the chips specify some maximum rise-time. A 1MOhm pull-up with a 50pF trace will already have an RC-time of 50microseconds. Quite a lot in digital communication to have to wait before you can safely use the same bus for another chip.
But, do you actually "need" them at all? Are you not driving the chip-selects with the chip directly? Do you not instruct the chip to drive them High when you're done talking? Then you don't need pull-ups.
If there's some more complex idea behind them you are welcome to share that for better answers, but just connecting a chip's output to another chip's input and driving that signal actively high or low is a very good reason to not use any resistor at all.**
And even if you do need a resistor, does it really need to be big? When you are communicating with the chips, how much power do they consume? More or less than 150uA? Most modules, not to mention your Atmel already waste miliamps to generate the serial signal, having a 10k pull-up at 1.8V only adds 0.18mA to that, which is very likely to be negligible. Since your pull up only has any current through it when the devices are talking, because when they are not, the line is high and the resistor will have 0V across it.
I = V/R = 0/whatever = 0
If you're actually asking about things like I2C pull-ups, not for hard driven CS signals, but for shared in/out lines such as SDA and SCL are, there's already many good questions and answers to them to browse through. For example, linked from this question at the time of asking (on the right):
**(apart from impedance matching series resistors, but that's 50 chapters ahead of this topic)