the scale on microcontrollers

the first MK gives 8bit, the second MK 8 bits, and latches the register.
I need to generate 40MHz (VGA)

It seems you need a video DAC (aka video encoder), it's about the way it works. For example, the THS8200. More options from Analog Devices.

But to generate the initial data flow is hardly possible to do with a set of small controllers, you also have the original data still forming, not just output. Although to display the statics of can and ride.
Better suited for this FPGA, the processor may generate the image and give it to Pliska, and that, in turn, is engaged exclusively in its conclusion. The data will be quite a lot, so you can store them in external RAM connected to the FPGA.

PS take a Look at the Altera DE-1 on this Board already have a simple R2R DAC on VGA output.

look in the direction i.MX23 Display Processing
in the processor built-in TV-Out VGA. the price of the processor is ~15$
(he yuzal i.mx27, works fine)

In theory you can, but there are a few problems:
— the microcontroller needs to do the operation in one clock cycle (about durakovich operations forget);
the results should be stored in a common register (this is a crutch);
— need a timer at 40MHz to actovate both controllers, switching cycles between the two microcontrollers (need some kind of crutch for the distribution of ticks);

ready decisions are not isolated... it is Easier to make a 16-bit microcontroller, 40MHz than marketing :)

10MHz controller on 20МИПС can be obtained only with great restrictions. And I'm not sure that each controller is able to give. For example, the AVR may each instruction to invert one output port, if it is to organize an infinite loop from such instructions without using branch instructions(it will abort the generation of this moment for a couple of cycles) we can talk about frequency 5 MHz on the output because each such instruction is executed over 2 cycles — and this is only half of the period of the generated signal. This controller will be engaged exclusively in the generation of a signal on one output and doing nothing else is able.

So the conclusion is — you need a controller soon. The alternatives are worse — in addition to combining multiple controllers in a "cluster" will also need a supervisor who will manage them synchronously, in General it is the work of making an elephant of a fly.
And at times, the normal controller can't do — you need hardware processing because the software to generate the video signal are not under force even the modern "big" CPUs for General use. There is another class of so-called video processors is a hybrid flexible programmable logic and cores in the form of a classical processor — this approach is used in modern graphics cards.

And for the formation of slowly changing images may be quite enough cords rigid logic + 2-port RAM (implemented in the usual way of demultiplexing bus address and data on two independent input / output) — single port, the read TSAP and the other port on the entry side of the controller. It usually all fits on a medium size FPGA-ine.