I have made many for friends.
Quality of prime components matter. L1 could be 100uH, but with 100uh marking we get 86uh only in market
C1 and c2 caps are to be close to. I used my box caps after precselection. and close match
A 1% error in 4mhz crystal cause 2% or more . Thus the crystal should be accurate enough.
Phil Rice had released a newer version.with sliced suffix. Try to us that firmware. He has also given facility of measuring F0 and F1 and gave clues for attaing those values.
The above guidelines could be of help to you.
One irritating issue is proximity of L1 to the relay child's magnetic field. Try to avoid that and keep L1 father.
All the best