what i misunderstand in dump's man:

     --kk _r_e_a_d _b_l_o_c_k_s_i_z_e
             The size in kilobyte of the read buffers, rounded up to a multi-
             ple of the filesystem block size. Default is 32k.


     --LL _l_a_b_e_l
             The user-supplied text string _l_a_b_e_l is placed into the dump head-
             er, where tools like restore(8) and file(1) can access it.  Note
             that this label is limited to be at most LBLSIZE (currently 16)
             characters, which must include the terminating `\0'.

     --nn      Whenever dduummpp requires operator attention, notify all operators
             in the group "operator" by means similar to a wall(1).

     --rr _c_a_c_h_e_s_i_z_e
             Use that many buffers for read cache operations.  A value of zero
             disables the read cache altogether, higher values improve read
             performance by reading larger data blocks from the disk and main-
             taining them in an LRU cache. See the --kk option for the size of
             the buffers. Maximum is 512, the size of the cache is limited to
             15% of the avail RAM by default.


i've started dump with

dump -0u -h 0 -B 5000000 -f /dev/rst0 -L "wojtek" -r 128 /

32k (default -k) *128 is 4MB

ps axv shows maximum 1184 total memory usage of one dump process.