rdma_performance用例use_rdma=false增加client端并发度qps反而降低

**Describe the bug**
将server和lclient部署在同一台服务器上做测试，rdma_performance用例中，使用use_rdma=false，走tcp路径，设置不同的并发度，发现qps并不是随着并发度的增加而增大的，在并发度2->3的时候，qps会出现很大幅度的下降

thread_num=1,
queue_depth=1: Avg-Latency: 19, 99th-Latency: 25, Throughput: 50.1614MB/s, QPS: 51k
queue_depth=2: Avg-Latency: 19, 99th-Latency: 34, Throughput: 98.1903MB/s, QPS: 100k
queue_depth=3: Avg-Latency: 72, 99th-Latency: 147, Throughput: 47.5866MB/s, QPS: 48k
queue_depth=4: Avg-Latency: 94, 99th-Latency: 165, Throughput: 40.9762MB/s, QPS: 41k
queue_depth=5: Avg-Latency: 103, 99th-Latency: 177, Throughput: 45.849MB/s, QPS: 46k
queue_depth=6: Avg-Latency: 90, 99th-Latency: 175, Throughput: 58.4754MB/s, QPS: 59k
queue_depth=7: Avg-Latency: 124, 99th-Latency: 202, Throughput: 54.4525MB/s, QPS: 55k
queue_depth=8: Avg-Latency: 99, 99th-Latency: 189, Throughput: 69.4976MB/s, QPS: 71k
queue_depth=9: Avg-Latency: 116, 99th-Latency: 190, Throughput: 71.0985MB/s, QPS: 72k
queue_depth=10: Avg-Latency: 113, 99th-Latency: 202, Throughput: 79.0347MB/s, QPS: 80k

queue_depth=1,
thread_num=1: Avg-Latency: 19, 99th-Latency: 25, Throughput: 50.1614MB/s, QPS: 51k
thread_num=2: Avg-Latency: 22, 99th-Latency: 36, Throughput: 87.3242MB/s, QPS: 89k
thread_num=3: Avg-Latency: 63, 99th-Latency: 140, Throughput: 46.3184MB/s, QPS: 47k
thread_num=4: Avg-Latency: 92, 99th-Latency: 164, Throughput: 40.8889MB/s, QPS: 41k
thread_num=5: Avg-Latency: 106, 99th-Latency: 189, Throughput: 45.1301MB/s, QPS: 46k
thread_num=6: Avg-Latency: 110, 99th-Latency: 188, Throughput: 51.0793MB/s, QPS: 52k
thread_num=7: Avg-Latency: 119, 99th-Latency: 201, Throughput: 54.6445MB/s, QPS: 55k
thread_num=8: Avg-Latency: 128, 99th-Latency: 206, Throughput: 59.5457MB/s, QPS: 60k
thread_num=9: Avg-Latency: 137, 99th-Latency: 221, Throughput: 63.7176MB/s, QPS: 65k
thread_num=10: Avg-Latency: 137, 99th-Latency: 224, Throughput: 69.4117MB/s, QPS: 71k

**To Reproduce**
server：./server --use_rdma=false
client：./client --use_rdma=false --attachment_size=1024 --echo_attachment=true --thread_num=1 --queue_depth=2
测试时改变client端的thread_num和queue_depth

**Expected behavior**
增加并发度qps应该有所增加，到达瓶颈之后应当维持不变，不应该出现下降的问题吧

**Versions**
OS: linux 6.6
Compiler: gcc 12.2
brpc: 1.15.0
protobuf: 28.3

**Additional context/screenshots**

Reproduce the issue with the provided server and client commands, using the rdma_performance TCP path with use_rdma=false. Vary client thread_num and queue_depth separately, record QPS and latency, and inspect the rdma_performance client/server entry points. Done means identifying and correcting the concurrency-related QPS drop, then confirming the results across the reported settings.

Written by the indexing model from the issue text.