process.cs

using System;
using System.Collections.Concurrent;
using System.Linq;
using System.Threading;
using System.Threading.Tasks;

// the task is to Analyze() files in parallel and Process() them on the main thread in batches
var files = Enumerable.Range(0, 1000).Select(x => $"File{x:0000}.cs").ToList();

var degreeOfParallelism = Math.Min(Environment.ProcessorCount, 8);
var unfinishedJob = new ConcurrentBag<string>(files);

var itemsToProcessBatch = 10;
var toProcess = new ConcurrentBag<string>();

while (unfinishedJob.Count > 0)
{
  // do Analyze() in parallel on background threads
  Task.WaitAll(Enumerable
    .Range(0, degreeOfParallelism)
    .Select(workerIndex =>
    {
      // can take some time, better to use long-running tasks?
      return Task.Run(() =>
      {
        while (unfinishedJob.TryTake(out var workItem))
        {
          if (Analyze(workerIndex, workItem))
          {
            toProcess.Add(workItem); // can actually add up to 20 items, that's fine
          }

          // this freezes the whole ConcurrentBag :(
          if (toProcess.Count > itemsToProcessBatch)
          {
            Console.Write('*');
            return; // time to process
          }
        }
      });
    })
    .ToArray());

  Console.WriteLine();

  // process the work items sequentially on the main thread
  while (toProcess.TryTake(out var workItem))
  {
    Process(workItem);
  }
}

bool Analyze(int workerIndex, string workItem)
{
  Thread.Sleep(100);
  var include = workItem.GetHashCode() % 20 == 0;
  Console.Write(include ? $"[{workerIndex}]" : workerIndex.ToString());
  return include; // 5%
}

void Process(string workItem)
{
  Console.WriteLine($"Processing {workItem}");
  Thread.Sleep(200);
}