It’s easier to group them like this when data is in a long format vice wide. Starting with your vectors:
DF2 <- data.frame(
x = c(c(A1, A2, A3), c(B1, B2, B3)),
y = rep(c("A", "B"), each = 15),
z = rep(rep(1:3, each=5), 2),
stringsAsFactors = FALSE
)
str(DF2)
# 'data.frame': 30 obs. of 3 variables:
# $ x: num 1 2 9 6 4 5 1 9 2 3 ...
# $ y: chr "A" "A" "A" "A" ...
# $ z: int 1 1 1 1 1 2 2 2 2 2 ...
cols <- rainbow(3, s = 0.5)
boxplot(x ~ z + y, data = DF2,
at = c(1:3, 5:7), col = cols,
names = c("", "A", "", "", "B", ""), xaxs = FALSE)
legend("topleft", fill = cols, legend = c(1,2,3), horiz = T)
The use of at manually controls the placement, so the “visual grouping” is not very robust. (You can control the spacing between them with width and/or boxwex.)