Maybe my estimations are wrong but...
According to the MSX mouse reading routine, the first long delay (that triggers the lecture sequence), is of 95667 ns (429 ticks with a Z80 at 3,58Mhz). The following delays are of 35011 ns (157 ticks), 37687 ns (169 ticks) and again 35011 ns (157 ticks).
But the Neos needs 135500 ns(542 ticks with a Z80 at 4Mhz), then 119250 ns (477 ticks), 114500 ns (458 ticks) and again 114500 ns (458 ticks).
I doubt a MSX mouse can work directly (with a Boxsoft interface) on an EP128 as only the reading routine needs 239 ticks each nibble....
On the other side, the Arduino(16 Mhz) adapter can be easily modified as it doesn't care about EP or MSX times...
------------------------------
I think, and possibly I'm wrong, that the delays of the Neos where stretched to fit the lower features of the C64 and other computers...