R Matrix class (not base matrix) indexing slow? [duplicate]

I am working with large matrices (~100k x 10k), some of which are dense, some of which are sparse. I am trying to use the Matrix class matrices as those implement a variety of types. However, one operation--selecting rows from matrices--seems to be orders of magnitude slower for dgeMatrix objects, the standard dense matrices in Matrix package:

library(Matrix)
library(microbenchmark)
A <- matrix(runif(10000*1000), 10000, 1000)
B <- Matrix(runif(10000*1000), 10000, 1000)
i <- 1234
j <- 987
microbenchmark(A[i,], A[,j], B[i,], B[,j])

This gives me

Unit: microseconds
   expr       min         lq        mean    median         uq       max neval
 A[i, ]    12.036    32.4635    48.46218    55.520    58.2860    96.707   100
 A[, j]    28.681    36.2045    41.96735    43.147    45.7985    68.484   100
 B[i, ] 35911.152 36444.2845 40367.11707 36637.658 37673.4705 63981.955   100
 B[, j] 36011.814 36417.5715 41399.72686 36731.589 37991.5010 60347.724   100

As you can see, the mean indexing of B (dgeMatrix) is almost 1000x slower than of A (base R matrix).

Am I doing something wrong? Am I using an inefficient class? Are there faster ways to mix sparse and dense matrices?