仿照网上写了一个方法 统计n个线程同时调一个方法，都执行完需要多长时间 ，发现一个奇怪的问题

package me.ele.coffe.hr.stress.test;  

public class test {  
    private static Integer error = 0;  
    private static Integer threads = 40;  
    private static Long startTime;
    private static Long endTime; 

    public static void main(String[] args) {  
        System.out.println("begin testing");
        System.out.println("线程数" + threads) ;
        WorkThread[] workThreads = new WorkThread[threads];  
        for (int i = 0; i < threads; i++) {  
            workThreads[i] = new WorkThread();  
        }  
        startTime = System.currentTimeMillis();  

        for (int i = 0; i < threads; i++) {  
            workThreads[i].start();  
        }  

    }  

    private static class WorkThread extends Thread {  
        public void run() {   
            try { 
               //要测试的函数 
                String s = testFunctoin.testString();
            } catch (Exception e) {  
                synchronized (error) {  
                    error++;  
                }  
                e.printStackTrace();  
            }  

            synchronized (threads) {  
                System.out.println("当前运行线程：====================" + getName());
                threads--;    
                if (threads == 0) { 
                    endTime = System.currentTimeMillis();
                    System.out.printf("总耗时：%d毫秒\n",endTime - startTime);  
                    System.out.printf("连接失败数量：%d\n", error);  
                }  
            }  
        }  

    }  
} 

发现当线程数较小时 可以打印出总耗时和连接失败数量
但当线程数较大时（比如40），就打印不出来了。。。
求帮忙看看 有没有哪里写的不对