By just adding the next predicate to avoid generation of immediate 16-bits stores
I got +5% jbb2005 on woodcrest:

  // Store Short/Char Immediate
  instruct storeImmI16(memory mem, immI16 src) %{
+ predicate(UseImm16ForStore);
    match(Set mem (StoreC mem src));

    ins_cost(150);
    format %{ "MOV16 $mem,$src" %}
    opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
    ins_encode( SizePrefix, OpcP, RMopc_Mem(0x00,mem), Con16( src ));
    ins_pipe( ialu_mem_imm );
  %}


hsdev-3% sh compare ref.jbb.n2 ref.jbb.n2_n16s
============================================================================
ref.jbb.n2
  Benchmark Samples Mean Stdev Geomean Weight
  specjbb2000 20 120380.43 591.43
  specjbb2005 20 51259.08 741.56
============================================================================
ref.jbb.n2_n16s
  Benchmark Samples Mean Stdev %Diff P Significant
  specjbb2000 20 122106.15 856.48 1.43 0.000 Yes
  specjbb2005 20 54137.34 989.22 5.62 0.000 Yes
============================================================================