Save the results of your rolling sum as something, then construct a vector containing the rows on which those sums ==4 are based (row, row+1)
rs2 <- with(df,rollapply(A, 2, sum, fill = NA))
s4 <- sort(unique(c(s4 <- which(rs2==4),s4+1)))
subset the appropriate rows
df[s4,]
# A b c
# 1 1 10 20
# 2 3 11 21
# 3 1 12 22
# 5 2 14 24
# 6 2 15 25
# 8 3 17 27
# 9 1 18 28