What is the fastest way to get a good conversion in screen 2 of a 8x8 image ?

I have a 8x8 RGB image at 24bits (8 bit for each R,G and B component)
and I need to convert it to screen 2 in an efficient way
I use the MSX 1 palette.

My current solution is "brute force"

1) I generate all the possible lines of 8 points that can be visualized by the VDP in screen 2 (excluding repetitions they are 26685)
2) I split the 8x8 block in 8 horizontal lines of 8 pixels
3) for each line, I compute the distance with the "allowed lines" and find the minimum (minimum squared error)

This algorithm implies that I compute 8 times 26685 distances between two 3D vectors (R,G,B components)

Moreover the result does not seem very nice, as I have seen that
some converters give very good results using dithering (e.g. Jannone's one, at msx.jannone.org, that seems excellent)

Any suggestion to get a faster/better conversion, maybe using dithering ?
Does any one have a snippet of code to share with me?

I think the results from this tool is very good.

*EDIT* Never mind, you already mentioned this tool

true, the problem is that the Floyd algorithm seems quite difficult
to implement...