Hi, with further encouragement, I decided to give tl(4) a try.
It's a very interesting ethernet chip, and even under heavy tcpbench
the mbuf usage is at most 6. As usual, feedback and comments are welcomed.

//Logan
C-x-C-c

Index: src/sys/dev/pci/if_tl.c
===================================================================
RCS file: /cvs/src/sys/dev/pci/if_tl.c,v
retrieving revision 1.50
diff -u -p -r1.50 if_tl.c
--- src/sys/dev/pci/if_tl.c	19 May 2010 15:27:35 -0000	1.50
+++ src/sys/dev/pci/if_tl.c	14 Dec 2010 19:15:28 -0000
@@ -252,8 +252,8 @@ int tl_intvec_rxeof(void *, u_int32_t);
 int tl_intvec_adchk(void *, u_int32_t);
 int tl_intvec_netsts(void *, u_int32_t);
 
-int tl_newbuf(struct tl_softc *,
-					struct tl_chain_onefrag *);
+int tl_newbuf(struct tl_softc *, struct tl_chain_onefrag *);
+void tl_fill_rx_ring(struct tl_softc *);
 void tl_stats_update(void *);
 int tl_encap(struct tl_softc *, struct tl_chain *,
 						struct mbuf *);
@@ -1034,8 +1034,6 @@ int tl_list_rx_init(sc)
 	for (i = 0; i < TL_RX_LIST_CNT; i++) {
 		cd->tl_rx_chain[i].tl_ptr =
 			(struct tl_list_onefrag *)&ld->tl_rx_list[i];
-		if (tl_newbuf(sc, &cd->tl_rx_chain[i]) == ENOBUFS)
-			return(ENOBUFS);
 		if (i == (TL_RX_LIST_CNT - 1)) {
 			cd->tl_rx_chain[i].tl_next = NULL;
 			ld->tl_rx_list[i].tlist_fptr = 0;
@@ -1047,34 +1045,45 @@ int tl_list_rx_init(sc)
 	}
 
 	cd->tl_rx_head = &cd->tl_rx_chain[0];
+	cd->tl_rx_cons = &cd->tl_rx_chain[0];
 	cd->tl_rx_tail = &cd->tl_rx_chain[TL_RX_LIST_CNT - 1];
-
+	cd->tl_rx_cnt = 0;
+	tl_fill_rx_ring(sc);
 	return(0);
 }
 
+void tl_fill_rx_ring(sc)
+	struct tl_softc         *sc;
+{
+	struct tl_list_data     *ld;
+	struct tl_chain_data    *cd;
+
+	cd = &sc->tl_cdata; 
+	ld = sc->tl_ldata;
+
+	while (cd->tl_rx_cnt < TL_RX_LIST_CNT) {
+		if(tl_newbuf(sc, cd->tl_rx_head) == ENOBUFS)
+			break;
+		cd->tl_rx_head = cd->tl_rx_head->tl_next;
+		cd->tl_rx_cnt++;   
+	}
+}       
+
 int tl_newbuf(sc, c)
 	struct tl_softc		*sc;
 	struct ...
[ message continues ]